Scaling of NAS BenchmarkNAS Class B Parallel Benchmark (v2.3)
Nasa Ames NAS parallel benchmarks
All rates in Mflops
.Benchmarks compiled with gnu g77 & MPICH.
Results in table are for 16 processor nodes only.
bt: linear equations for implicit scheme in Navier-Stokes equation
is: integer sort
cg: conjugate gradient matrix inversion
ep: parallel random numbers
lu: LU decomposition for Navier-Stokes equation
mg: multigrid code fragment
sp: linear equations for Navier-Stokes equation
|
|
Aeneas |
Cray T3E-1200 |
Cray Origin 2000 |
IBM SP2 (160MHz) |
|
bt |
535.65 |
1054.8 |
890.9 |
1996.4 |
|
is |
6.47 |
37.7 |
30.2 |
37.2 |
|
cg |
124.40 |
183.7 |
428.4 |
446.8 |
|
ep |
11.06 |
54.5 |
68.8 |
36.5 |
|
lu |
642.09 |
1144.9 |
1382.5 |
1956.9 |
|
mg |
288.07 |
1508.7 |
1035.2 |
1992.0 |
|
sp |
276.98 |
784.0 |
617.8 |
1411.2 |
Compiler switches: g77 –static –O3 –m486 –funroll-loops –malign-double –ffast-math
Scaling of NAS Benchmark
Performance of bt,cg,ep,is,lu,mg and sp code fragments on 1-16 processor nodes
For a complete benchmark description, see
http://www.nas.nasa.gov/benchmarks/npb2.3.html
5/14/1998